[AusNOG] MPLS/VPLS solution

[Matthew Moyle-Croft]

Lincoln,
The optical switches still exist for 10GE customers.   They pretty much reused their existing Brocade platform except for larger core switches (what was possible changed when they moved to VPLS).

I'd forgotten about the VSRP stuff - Henk's preso from APRICOT2010 explains it all:

http://meetings.apnic.net/__data/assets/pdf_file/0010/18856/APOPS-Plenary-2_AMS-IX-Generation-4-MPLSVPLS-experiences-_Henk-Steenman.pdf

Certainly the migration was fairly smooth - we didn't notice really.

MMC


On 10/09/2010, at 12:40 PM, Lincoln Dale wrote:


On 10/09/2010, at 12:31 PM, Matthew Moyle-Croft wrote:

One of the more interesting applications I've seen of VPLS is for building big IXes.

AMS-IX has migrated entirely to a VPLS network.   It appears to be much more stable and resilient than a spanning-tree based network.  Given that AMS-IX peaks at around 1Tbps of traffic and has 649 or so ports - it's a pretty good indication of the scale you can go for a single domain.


AMS-IX wasn't STP based - well - it was once, but we're talking 2003/2004 timeframe, i.e. not for ages.

prior to migrating to VPLS, they used a combination of optical cross-connect (Glimmerglass photonic switch) and Foundry/Brocade's proprietary VSRP which effectively meant they built out redundant A/B networks and optically switched between the two.
migrating to VPLS solved some of the unique scaling challenges that they had with the architecture they had.


TRILL/Rbridges would have been ideal for AMS-IX but for the fact that it wasn't yet close enough in terms of shipping product to meet their required timeframes given the continued growth trajectory of traffic & 10G ports.
likewise the optical transceiver form factor preferred for large-port-count 10G switches (SFP+) don't yet provide 10G DWDM connectivity which they also required since they are spread out over multiple physical locations with limited dark fiber between sites.

business/timing reasons aside, it is a demonstration of how a single L2 domain can scale to Tbps of traffic.


Admittedly IXes are a bit special in terms of needs (ie. single mac per customer with filters).

certainly there are unique requirements to IXes.



cheers,

lincoln.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20100910/e542ba5a/attachment.html>